CN104750674B - A kind of man-machine conversation's satisfaction degree estimation method and system - Google Patents
A kind of man-machine conversation's satisfaction degree estimation method and system Download PDFInfo
- Publication number
- CN104750674B CN104750674B CN201510086568.4A CN201510086568A CN104750674B CN 104750674 B CN104750674 B CN 104750674B CN 201510086568 A CN201510086568 A CN 201510086568A CN 104750674 B CN104750674 B CN 104750674B
- Authority
- CN
- China
- Prior art keywords
- man
- machine conversation
- training
- session characteristics
- satisfaction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention discloses a kind of man-machine conversation's satisfaction degree estimation method and system, and method includes: the satisfaction training data for obtaining multiple trained man-machine conversations；To train the session characteristics of man-machine conversation as training sample, disaggregated model training is carried out using corresponding satisfaction result as target value, obtains disaggregated model；Practical man-machine conversation is obtained, the session characteristics of the practical man-machine conversation are extracted, the session characteristics of the practical man-machine conversation are inputted into the disaggregated model and obtain the satisfaction result predicted by disaggregated model.The present invention is when predicting satisfaction, it can be easy to add or delete the session characteristics for influencing prediction, and pass through statistical, it no longer needs directly to consider the weight that each feature specifically needs to set, and it only needs to select suitable disaggregated model, study using disaggregated model to training sample, is automatically performed weight setting.
Description
Technical field
The present invention relates to especially a kind of man-machine conversation's satisfaction degree estimation method and system of man-machine conversation's correlative technology field.
Background technique
Question answering system is the natural language text sentence that can identify user's input, and makes the computer journey accordingly answered
Sequence.It is analyzed by the processing to user's read statement, final to execute user's request, the information that user is concerned about is returned to, wherein wrapping
Containing natural language processings the relevant technologies such as sentence participle, Entity recognition, semantics recognition, task processing and responses.
In question answering system, the accuracy of answer, real-time, pleasantly surprised property etc. is all important index, and improves these
The final purpose of performance is provided to serve client, allows customer satisfaction.In order to preferably serve customer service, need to know assorted
The user of sample when can satisfaction, thus can targetedly to may dissatisfaction client carry out volume
Outer operation.
Existing satisfaction degree estimation is mainly predicted by the manually mode of laying down a regulation, for example is occurred in user's question
Negative emotions, then it is assumed that the possible dissatisfaction of client, for another example client was to like playing satisfaction customer service in the past, then it is assumed that client will not
Then short shot meaning degree etc. formulates weight to various modes, then weighting finds out last satisfaction degree estimation situation.
However, being then weighted ballot is a kind of more original processing mode by laying down a regulation, need a large amount of
It tests to adjust the weights of various rules, if rule type is excessive, and when certain association is individually present, artificial treatment is multiple
Miscellaneous degree can growth at double, it is also not convenient for safeguarding；Some pairs of satisfactions are not the rules directly affected, are not allowed easy to handle yet.
Summary of the invention
Based on this, it is necessary to for the prior art pair technical problem complex with the satisfaction degree estimation of client, provide
A kind of man-machine conversation's satisfaction degree estimation method and system.
A kind of man-machine conversation's satisfaction degree estimation method, comprising:
Training data collection step, comprising: the satisfaction training data of multiple trained man-machine conversations is obtained, it is each described full
Meaning degree training data includes user about the satisfaction result of this time training man-machine conversation and in this time training man-machine conversation
Extracted session characteristics；
Corpus training step, comprising: to train the session characteristics of man-machine conversation as training sample, with corresponding satisfaction
As a result disaggregated model training is carried out as target value, obtains disaggregated model, the disaggregated model is by the session as training sample
Feature carries out classification with the satisfaction result as target value and is associated with；
Satisfaction degree estimation step, comprising: practical man-machine conversation is obtained, the session characteristics of the practical man-machine conversation are extracted,
The session characteristics of the practical man-machine conversation are inputted into the disaggregated model and obtain the satisfaction result predicted by disaggregated model.
A kind of man-machine conversation's satisfaction degree estimation system, comprising:
Training data collection module, is used for: the satisfaction training data of multiple trained man-machine conversations is obtained, it is each described full
Meaning degree training data includes user about the satisfaction result of this time training man-machine conversation and in this time training man-machine conversation
Extracted session characteristics；
Corpus training module, is used for: to train the session characteristics of man-machine conversation as training sample, with corresponding satisfaction
As a result disaggregated model training is carried out as target value, obtains disaggregated model, the disaggregated model is by the session as training sample
Feature carries out classification with the satisfaction result as target value and is associated with；
Satisfaction degree estimation module, is used for: practical man-machine conversation obtained, the session characteristics of the practical man-machine conversation are extracted,
The session characteristics of the practical man-machine conversation are inputted into the disaggregated model and obtain the satisfaction result predicted by disaggregated model.
The present invention is pre- to practical man-machine conversation progress satisfaction using disaggregated model by being trained to disaggregated model
It surveys, when predicting satisfaction, can be easy to add or delete the session characteristics for influencing prediction, and by statistical, no
It needs directly to consider the weight that each feature specifically needs to set again, and only needs to select suitable disaggregated model, utilize classification
Study of the model to training sample, is automatically performed weight setting.
Detailed description of the invention
Fig. 1 is a kind of work flow diagram of man-machine conversation's satisfaction degree estimation method of the present invention；
Fig. 2 is a kind of structure chart of man-machine conversation's satisfaction degree estimation system of the present invention.
Specific embodiment
The present invention will be further described in detail in the following with reference to the drawings and specific embodiments.
It is as shown in Figure 1 a kind of work flow diagram of man-machine conversation's satisfaction degree estimation method of the present invention, comprising:
Step S101, comprising: obtain the satisfaction training data of multiple trained man-machine conversations, each satisfaction training
Data include that user trains the satisfaction result of man-machine conversation about this time and trains in this time extracted in man-machine conversation
Session characteristics；
Step S102, comprising: to train the session characteristics of man-machine conversation as training sample, with corresponding satisfaction result
Disaggregated model training is carried out as target value, obtains disaggregated model, the disaggregated model is by the session characteristics as training sample
Classification is carried out with the satisfaction result as target value to be associated with；
Step S103, comprising: practical man-machine conversation is obtained, the session characteristics of the practical man-machine conversation are extracted, it will be described
The session characteristics of practical man-machine conversation input the disaggregated model and obtain the satisfaction result predicted by disaggregated model.
The present invention obtains multiple satisfaction training datas, and the feature extraction that conversates in step s101, in step
Disaggregated model training is carried out in S102 using session characteristics as training sample, obtains disaggregated model.Disaggregated model includes that logic is returned
Return, the models such as vector machine (Support Vector Machine, SVM), random forest, is that example is assigned to by various features
The model of one known classification.In step s 103, session characteristics are extracted from the practical man-machine conversation got, and inputted
The obtained disaggregated model of step S102 training, thus the satisfaction result after being predicted.
The present invention can be easy to add or delete the session characteristics for influencing prediction, and by statistical, no longer need
Directly to consider the weight that each feature specifically needs to set, and only need to select suitable disaggregated model, utilize disaggregated model
Study to training sample, is automatically performed weight setting.
When obtain by disaggregated model predict satisfaction result after, can use the satisfaction as a result, adjusting man-machine meeting
Strategy is answered by robot in words, or is alerted to prompt the artificial customer service in backstage to follow up.
In one of the embodiments, further include:
Alarm step, comprising: if meeting preset warning condition by the satisfaction result that disaggregated model is predicted, to
The artificial customer service alarm device docked with the man-machine conversation issues warning information.
In the present embodiment, when satisfaction result meets preset warning condition, then issued in artificial customer service alarm device
Warning information reminds artificial customer service auxiliary robot services client.Warning condition is configured according to satisfaction result, such as such as
Fruit satisfaction result is only satisfied or is unsatisfied with the two, then warning condition are as follows: and " satisfaction result is dissatisfied ", if satisfaction
As a result using star identify, such as it is most satisfied be five-pointed star, satisfaction be four stars, generally Samsung, be discontented with mean two stars, least be satisfied with
For a star, then warning condition can be with are as follows: " satisfaction result is less than Samsung ".Specific setting those of ordinary skill in the art are readding
It can be arranged according to the actual situation after reader patent.
Session characteristics can be with are as follows: the type of customer problem, user sources entrance, user seek advice from Taxonomy Information etc..
In one of the embodiments:
The trained man-machine conversation includes at least one training user's problem, the extracted session in training man-machine conversation
Feature includes: the Questions types as made of training user's problem conversion；
The practical man-machine conversation includes at least one actual user's problem, the extracted session in practical man-machine conversation
Feature includes: the Questions types as made of actual user's problem conversion.
The list of Questions types is converted to for training user's problem as shown in table 1, actual user's problem, which is converted to, asks
Topic type can be realized using same or similar mode.
| SessionID | Customer problem | Questions types | User sources entrance | …… | Satisfactory result |
| Session1 | How this commodity is out of stock | Commodity stocks inquiry | Online | It is satisfied | |
| Session1 | When arrival | It is when available in stock | Online | ||
| Session2 | I has descended an order | Order inquiries | Other | It is dissatisfied | |
| Session2 | When arrival | Distribution time | Other | ||
| Session3 | You are good | Greeting | Other | It is satisfied | |
| Session3 | When available in stock this commodity is | It is when available in stock | Other |
In table 1, SessionID is for identifying trained man-machine conversation, and identical trained man-machine conversation is using same
SessionID.Customer problem can extract keyword therein using existing semantic analysis and word division methods, thus will
It is referred in suitable Questions types.
In man-machine conversation, user proposes various problems to robot, and robot makes a response according to problem, automatic to carry out
It answers.The problem of different type, the satisfaction of user is different, for example, for when it is available in stock such issues that, due to machine
People can obtain information from the deposit system of storehouse, therefore can answer customer problem well, therefore be easy that user is allowed to please oneself, and
For other problems, robot may may not can obtain accurate answer or robot may not can accurate understanding user ask
Topic, such situation are easier to that user is allowed to feel discontented.Therefore the session characteristics of user are classified as Questions types, and root by the present invention
The satisfaction of user is judged according to Questions types.
In one of the embodiments:
The extracted session characteristics in training man-machine conversation further include: the trained man-machine conversation's carrys out source inlet；
The extracted session characteristics in practical man-machine conversation further include: the practical man-machine conversation's carrys out source inlet.
Example as shown in table 1, the different source inlets that comes will lead to the difference of information acquired in robot, therefore, by not
The same source inlet that comes distinguishes the accuracy that can increase satisfaction degree estimation.
In one of the embodiments:
Feature vector will be converted to by extracted session characteristics in training man-machine conversation, in the step S102, with instruction
The corresponding feature vector of session characteristics for practicing man-machine conversation carries out disaggregated model training, described eigenvector packet as training sample
It includes and enters with the source of the extracted corresponding Questions types of session characteristics and the trained man-machine conversation in training man-machine conversation
The one-to-one position of mouth, and it is corresponding with the corresponding Questions types of session characteristics included by training man-machine conversation in feature vector
Position position, come the corresponding position position of source inlet with training man-machine conversation in feature vector；
Feature vector will be converted to by extracted session characteristics in practical man-machine conversation, in the step S103, with reality
The corresponding feature vector of the session characteristics of border man-machine conversation inputs the disaggregated model and obtains the satisfaction predicted by disaggregated model
Degree is as a result, described eigenvector includes Questions types corresponding with session characteristics extracted in practical man-machine conversation and described
Practical man-machine conversation's comes the one-to-one position of source inlet, and in feature vector with session characteristics included by practical man-machine conversation
The corresponding position of source inlet is come with practical man-machine conversation in feature vector in corresponding Questions types corresponding position position
Position.
Session characteristics in same trained man-machine conversation or same practical man-machine conversation are merged into a feature vector,
Consequently facilitating the training and calculating of subsequent classification model.
In one of the embodiments:
The extracted session characteristics in training man-machine conversation further include: the satisfaction training data corresponds to user
Historical behavior feature；
The session characteristics extracted in practical man-machine conversation further include: the practical man-machine conversation corresponds to user's
Historical behavior feature.
The historical behavior feature of user refers to user in the man-machine conversation of history to the history evaluation of satisfaction result.
For example, if the satisfaction result majority of user be all it is satisfied, then same user is for the satisfaction in practical man-machine conversation
As a result evaluation remains as satisfied probability with regard to larger.
In one of the embodiments: the man-machine conversation is purchase system man-machine conversation；
The extracted session characteristics in training man-machine conversation further include: the satisfaction training data corresponds to user
Shopping status；
The session characteristics extracted in practical man-machine conversation further include: the practical man-machine conversation corresponds to user's
Shopping status.
In shopping process, the different shopping status of user can have large effect to its satisfaction, such as user is under
It is then easy to be discontented with if it is inquiry distribution time after list, and is easier to before placing an order if only when available in stock inquiry is
It is satisfied.
As highly preferred embodiment of the present invention, made with the Liblinear that Chih-Jen doctor Lin of Taiwan Univ. develops
For disaggregated model tool, wherein solverType (algorithm types) selects L2R_LR (the normalized logistic regression algorithm of L2), C
(penalty factor) is set as 4.0, eps (iteration stopping threshold value) and is set as 0.01.
Wherein, data required for Liblinear training pattern input as a two-dimensional matrix, wherein every a line is one
A sample vector, each column represent a feature, indicate that sample is all corresponding with one for each sample in the value of this feature
A target value.
All input datas are trained to a model file by Liblinear, when needing to identify, input one
A sample vector, so that it may predict its possible target value.
For satisfaction degree estimation, defining each logical session (including multiple question and answer) is a training sample, in multiple question and answer
All features are merged into a feature vector, and user's authentic assessment result is as target value, for example, taking for the data of JIMI
Data training in one month, extracts the features such as Questions types, entrance source, user gradation, if there is specific characteristic, characteristic value
1 is taken, for multivalue discrete features, resolves into multiple features, that Discrete Eigenvalue of appearance takes 1, for the spy not occurred
Sign does not have to processing.
As shown in table 1, the two-dimensional matrix sample after conversion is as shown in table 2 for dialogue, reads corpus by program, and be converted to
Liblinear need input format, formed training set training, obtain disaggregated model, then user come JIMI put question to when into
Row prediction.
Table 2
By using disaggregated model, theoretical foundation is provided for whether prediction user session is satisfied with, and reduce people
Work extracts, analyzes feature importance, and sets the time of weight, and when colleague has found useful attribute, addition verifying is also easy to.Have
Satisfaction degree estimation, it is easier to grasp robot service user situation, and can make timely adjustment as the case may be, meet
Different user demands.
It is illustrated in figure 2 a kind of structure chart of man-machine conversation's satisfaction degree estimation system of the present invention, comprising:
Training data collection module 201, is used for: obtaining the satisfaction training data of multiple trained man-machine conversations, Mei Gesuo
Stating satisfaction training data includes user about the satisfaction result of this time training man-machine conversation and in the secondary training of human chance
Extracted session characteristics in words；
Corpus training module 202, is used for: to train the session characteristics of man-machine conversation as training sample, with corresponding full
Meaning degree result carries out disaggregated model training as target value, obtains disaggregated model, the disaggregated model will be as training sample
Session characteristics carry out classification with the satisfaction result as target value and are associated with；
Satisfaction degree estimation module 203, is used for: obtaining practical man-machine conversation, the session for extracting the practical man-machine conversation is special
The session characteristics of the practical man-machine conversation are inputted the disaggregated model and obtain the satisfaction knot predicted by disaggregated model by sign
Fruit.
In one of the embodiments, further include:
Alarm module is used for: if meeting preset warning condition by the satisfaction result that disaggregated model is predicted, to
The artificial customer service alarm device docked with the man-machine conversation issues warning information.
In one of the embodiments:
The trained man-machine conversation includes at least one training user's problem, the extracted session in training man-machine conversation
Feature includes: the Questions types as made of training user's problem conversion；
The practical man-machine conversation includes at least one actual user's problem, the extracted session in practical man-machine conversation
Feature includes: the Questions types as made of actual user's problem conversion.
In one of the embodiments:
The extracted session characteristics in training man-machine conversation further include: the trained man-machine conversation's carrys out source inlet；
The extracted session characteristics in practical man-machine conversation further include: the practical man-machine conversation's carrys out source inlet.
In one of the embodiments:
Feature vector will be converted to by extracted session characteristics in training man-machine conversation, in the corpus training step,
Using train the corresponding feature vector of session characteristics of man-machine conversation as training sample carry out disaggregated model training, the feature to
Amount include and training man-machine conversation in the extracted corresponding Questions types of session characteristics and the trained man-machine conversation come
The one-to-one position of source inlet, and Questions types phase corresponding with session characteristics included by training man-machine conversation in feature vector
The corresponding position position of source inlet is come with training man-machine conversation in feature vector in corresponding position position；
Feature vector, the satisfaction degree estimation step will be converted to by extracted session characteristics in practical man-machine conversation
In, the disaggregated model is inputted with the corresponding feature vector of the session characteristics of practical man-machine conversation and obtains predicting by disaggregated model
Satisfaction as a result, described eigenvector include Questions types corresponding with session characteristics extracted in practical man-machine conversation with
And the practical man-machine conversation comes the one-to-one position of source inlet, and in feature vector with meeting included by practical man-machine conversation
Talk about the corresponding position position of the corresponding Questions types of feature, in feature vector with practical man-machine conversation to carry out source inlet corresponding
Position position.
In one of the embodiments:
The extracted session characteristics in training man-machine conversation further include: the satisfaction training data corresponds to user
Historical behavior feature；
The session characteristics extracted in practical man-machine conversation further include: the practical man-machine conversation corresponds to user's
Historical behavior feature.
In one of the embodiments: the man-machine conversation is purchase system man-machine conversation；
The extracted session characteristics in training man-machine conversation further include: the satisfaction training data corresponds to user
Shopping status；
The session characteristics extracted in practical man-machine conversation further include: the practical man-machine conversation corresponds to user's
Shopping status.
The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously
Limitations on the scope of the patent of the present invention therefore cannot be interpreted as.It should be pointed out that for those of ordinary skill in the art
For, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to guarantor of the invention
Protect range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (8)
1. a kind of man-machine conversation's satisfaction degree estimation method characterized by comprising
Training data collection step, comprising: obtain the satisfaction training data of multiple trained man-machine conversations, each satisfaction
Training data includes that user is mentioned about the satisfaction result of this time training man-machine conversation and in this time training man-machine conversation
The session characteristics taken；
Corpus training step, comprising: to train the session characteristics of man-machine conversation as training sample, with corresponding satisfaction result
Disaggregated model training is carried out as target value, obtains disaggregated model, the disaggregated model is by the session characteristics as training sample
Classification is carried out with the satisfaction result as target value to be associated with；
Satisfaction degree estimation step, comprising: obtain practical man-machine conversation, the session characteristics of the practical man-machine conversation are extracted, by institute
The session characteristics for stating practical man-machine conversation input the disaggregated model and obtain the satisfaction result predicted by disaggregated model；
The trained man-machine conversation includes at least one training user's problem, the extracted session characteristics in training man-machine conversation
It include: the Questions types as made of training user's problem conversion；
The practical man-machine conversation includes at least one actual user's problem, the extracted session characteristics in practical man-machine conversation
It include: the Questions types as made of actual user's problem conversion；
The extracted session characteristics in training man-machine conversation further include: the trained man-machine conversation's carrys out source inlet；
The extracted session characteristics in practical man-machine conversation further include: the practical man-machine conversation's carrys out source inlet；
The extracted session characteristics in training man-machine conversation further include: the satisfaction training data corresponds to going through for user
History behavioural characteristic；
The session characteristics extracted in practical man-machine conversation further include: the practical man-machine conversation corresponds to the history of user
Behavioural characteristic.
2. man-machine conversation's satisfaction degree estimation method according to claim 1, which is characterized in that further include:
Alarm step, comprising: if by disaggregated model predict satisfaction result meet preset warning condition, to institute
The artificial customer service alarm device for stating man-machine conversation's docking issues warning information.
3. man-machine conversation's satisfaction degree estimation method according to claim 1, it is characterised in that:
Feature vector will be converted to by extracted session characteristics in training man-machine conversation, in the corpus training step, with instruction
The corresponding feature vector of session characteristics for practicing man-machine conversation carries out disaggregated model training, described eigenvector packet as training sample
It includes and enters with the source of the extracted corresponding Questions types of session characteristics and the trained man-machine conversation in training man-machine conversation
The one-to-one position of mouth, and it is corresponding with the corresponding Questions types of session characteristics included by training man-machine conversation in feature vector
Position position, come the corresponding position position of source inlet with training man-machine conversation in feature vector；
Feature vector will be converted to by extracted session characteristics in practical man-machine conversation, in the satisfaction degree estimation step, with
The corresponding feature vector of the session characteristics of practical man-machine conversation inputs the disaggregated model and obtains expiring by what disaggregated model was predicted
Meaning degree is as a result, described eigenvector includes Questions types corresponding with session characteristics extracted in practical man-machine conversation and institute
That states practical man-machine conversation comes the one-to-one position of source inlet, and special with session included by practical man-machine conversation in feature vector
It levies the corresponding position position of corresponding Questions types, comes the corresponding position of source inlet with practical man-machine conversation in feature vector
Position.
4. man-machine conversation's satisfaction degree estimation method according to claim 1, it is characterised in that: the man-machine conversation is shopping
System man-machine conversation；
The extracted session characteristics in training man-machine conversation further include: the satisfaction training data corresponds to the purchase of user
Object state；
The session characteristics extracted in practical man-machine conversation further include: the practical man-machine conversation corresponds to the shopping of user
State.
5. a kind of man-machine conversation's satisfaction degree estimation system characterized by comprising
Training data collection module, is used for: obtaining the satisfaction training data of multiple trained man-machine conversations, each satisfaction
Training data includes that user is mentioned about the satisfaction result of this time training man-machine conversation and in this time training man-machine conversation
The session characteristics taken；
Corpus training module, is used for: to train the session characteristics of man-machine conversation as training sample, with corresponding satisfaction result
Disaggregated model training is carried out as target value, obtains disaggregated model, the disaggregated model is by the session characteristics as training sample
Classification is carried out with the satisfaction result as target value to be associated with；
Satisfaction degree estimation module, is used for: obtaining practical man-machine conversation, the session characteristics of the practical man-machine conversation is extracted, by institute
The session characteristics for stating practical man-machine conversation input the disaggregated model and obtain the satisfaction result predicted by disaggregated model；
The trained man-machine conversation includes at least one training user's problem, the extracted session characteristics in training man-machine conversation
It include: the Questions types as made of training user's problem conversion；
The practical man-machine conversation includes at least one actual user's problem, the extracted session characteristics in practical man-machine conversation
It include: the Questions types as made of actual user's problem conversion；
The extracted session characteristics in training man-machine conversation further include: the trained man-machine conversation's carrys out source inlet；
The extracted session characteristics in practical man-machine conversation further include: the practical man-machine conversation's carrys out source inlet；
The extracted session characteristics in training man-machine conversation further include: the satisfaction training data corresponds to going through for user
History behavioural characteristic；
The session characteristics extracted in practical man-machine conversation further include: the practical man-machine conversation corresponds to the history of user
Behavioural characteristic.
6. man-machine conversation's satisfaction degree estimation system according to claim 5, which is characterized in that further include:
Alarm module is used for: if by disaggregated model predict satisfaction result meet preset warning condition, to institute
The artificial customer service alarm device for stating man-machine conversation's docking issues warning information.
7. man-machine conversation's satisfaction degree estimation system according to claim 5, it is characterised in that:
Feature vector will be converted to by extracted session characteristics in training man-machine conversation, in the corpus training step, with instruction
The corresponding feature vector of session characteristics for practicing man-machine conversation carries out disaggregated model training, described eigenvector packet as training sample
It includes and enters with the source of the extracted corresponding Questions types of session characteristics and the trained man-machine conversation in training man-machine conversation
The one-to-one position of mouth, and it is corresponding with the corresponding Questions types of session characteristics included by training man-machine conversation in feature vector
Position position, come the corresponding position position of source inlet with training man-machine conversation in feature vector；
Feature vector will be converted to by extracted session characteristics in practical man-machine conversation, in the satisfaction degree estimation step, with
The corresponding feature vector of the session characteristics of practical man-machine conversation inputs the disaggregated model and obtains expiring by what disaggregated model was predicted
Meaning degree is as a result, described eigenvector includes Questions types corresponding with session characteristics extracted in practical man-machine conversation and institute
That states practical man-machine conversation comes the one-to-one position of source inlet, and special with session included by practical man-machine conversation in feature vector
It levies the corresponding position position of corresponding Questions types, comes the corresponding position of source inlet with practical man-machine conversation in feature vector
Position.
8. man-machine conversation's satisfaction degree estimation system according to claim 5, it is characterised in that: the man-machine conversation is shopping
System man-machine conversation；
The extracted session characteristics in training man-machine conversation further include: the satisfaction training data corresponds to the purchase of user
Object state；
The session characteristics extracted in practical man-machine conversation further include: the practical man-machine conversation corresponds to the shopping of user
State.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510086568.4A CN104750674B (en) | 2015-02-17 | 2015-02-17 | A kind of man-machine conversation's satisfaction degree estimation method and system |
| HK15109493.6A HK1208931A1 (en) | 2015-02-17 | 2015-09-25 | Method and system for predicting satisfaction of a human-machine conversation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510086568.4A CN104750674B (en) | 2015-02-17 | 2015-02-17 | A kind of man-machine conversation's satisfaction degree estimation method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104750674A CN104750674A (en) | 2015-07-01 |
| CN104750674B true CN104750674B (en) | 2018-12-21 |
Family
ID=53590386
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510086568.4A Active CN104750674B (en) | 2015-02-17 | 2015-02-17 | A kind of man-machine conversation's satisfaction degree estimation method and system |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN104750674B (en) |
| HK (1) | HK1208931A1 (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104951434B (en) * | 2015-07-10 | 2018-10-23 | 沃民高新科技（北京）股份有限公司 | The determination method and apparatus of brand mood |
| CN105095415B (en) * | 2015-07-10 | 2018-12-21 | 沃民高新科技（北京）股份有限公司 | The determination method and apparatus of network mood |
| CN104951807B (en) * | 2015-07-10 | 2018-09-25 | 沃民高新科技（北京）股份有限公司 | The determination method and apparatus of stock market's mood |
| CN105590225A (en) * | 2015-12-22 | 2016-05-18 | 成都小多科技有限公司 | Sales stage identification method and device based on client problems |
| CN105654250A (en) * | 2016-02-01 | 2016-06-08 | 百度在线网络技术（北京）有限公司 | Method and device for automatically assessing satisfaction degree |
| CN105701088B (en) * | 2016-02-26 | 2018-12-28 | 北京京东尚科信息技术有限公司 | The method and apparatus manually talked with are switched to from machine dialogue |
| CN105808652B (en) * | 2016-02-26 | 2021-05-25 | 北京京东尚科信息技术有限公司 | Method and device for realizing online customer service |
| CN105760965A (en) * | 2016-03-15 | 2016-07-13 | 北京百度网讯科技有限公司 | Pre-estimated model parameter training method, service quality pre-estimation method and corresponding devices |
| CN106844750A (en) * | 2017-02-16 | 2017-06-13 | 深圳追科技有限公司 | Emotion is pacified in a kind of robot based on customer service man-machine interaction method and system |
| CN107463301A (en) * | 2017-06-28 | 2017-12-12 | 北京百度网讯科技有限公司 | Conversational system construction method, device, equipment and computer-readable recording medium based on artificial intelligence |
| CN110019710A (en) * | 2017-11-27 | 2019-07-16 | 厦门快商通信息技术有限公司 | A kind of topic forest formula interactive method and system |
| CN112417842A (en) * | 2019-08-19 | 2021-02-26 | 阿里巴巴集团控股有限公司 | Service satisfaction determining method and device, storage medium and computer equipment |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101394316B (en) * | 2008-11-11 | 2011-08-31 | 南京大学 | Network intrusion classification method based on Bayes of total indirected graph |
| CN102937951B (en) * | 2011-08-15 | 2016-11-02 | 北京百度网讯科技有限公司 | Set up the method for IP address sort model, the method and device to user's classification |
| WO2014069121A1 (en) * | 2012-10-31 | 2014-05-08 | 日本電気株式会社 | Conversation analysis device and conversation analysis method |
| CN103067623A (en) * | 2013-01-29 | 2013-04-24 | 江苏电力信息技术有限公司 | Telephone traffic peak surge calling optimization method based on collaborative filtering |
| CN103996130B (en) * | 2014-04-29 | 2016-04-27 | 北京京东尚科信息技术有限公司 | A kind of information on commodity comment filter method and system |
-
2015
- 2015-02-17 CN CN201510086568.4A patent/CN104750674B/en active Active
- 2015-09-25 HK HK15109493.6A patent/HK1208931A1/en unknown
Non-Patent Citations (2)
| Title |
|---|
| 人机对话识别系统在移动领域的应用实践;王功勋;《中国优秀硕士学位论文全文数据库 信息科技辑》;20130228;正文第32-39页 * |
| 基于句类模型的人机对话系统语句库设计;梁子琳;《华中科技大学学报（自然科学版）》;20111130;第316-319页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| HK1208931A1 (en) | 2016-03-18 |
| CN104750674A (en) | 2015-07-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104750674B (en) | A kind of man-machine conversation's satisfaction degree estimation method and system | |
| CN109902222B (en) | Recommendation method and device | |
| CN104809103B (en) | A kind of interactive semantic analysis and system | |
| CN109816483B (en) | Information recommendation method and device and readable storage medium | |
| CN109325691A (en) | Abnormal behaviour analysis method, electronic equipment and computer program product | |
| JP2021504789A (en) | ESG-based corporate evaluation execution device and its operation method | |
| CN106649742A (en) | Database maintenance method and device | |
| CN107506350B (en) | Method and equipment for identifying information | |
| CN109978020B (en) | Social network account number vest identity identification method based on multi-dimensional features | |
| CN110347833B (en) | Classification method for multi-round conversations | |
| CN107730131A (en) | The ability prediction of mass-rent software developer a kind of and recommendation method, apparatus | |
| CN110827797B (en) | Voice response event classification processing method and device | |
| WO2017071474A1 (en) | Method and device for processing language data items and method and device for analyzing language data items | |
| CN108038208A (en) | Training method, device and the storage medium of contextual information identification model | |
| CN113297365B (en) | User intention judging method, device, equipment and storage medium | |
| CN109960791A (en) | Judge the method and storage medium, terminal of text emotion | |
| CN114037545A (en) | Client recommendation method, device, equipment and storage medium | |
| CN112348417A (en) | Marketing value evaluation method and device based on principal component analysis algorithm | |
| CN115909443A (en) | Expression recognition model based on multi-head attention mechanism and training method thereof | |
| Costa et al. | Automatic classification of computational thinking skills in elementary school math questions | |
| CN103886869B (en) | A kind of information feedback method based on speech emotion recognition and system | |
| CN109190116A (en) | Semantic analytic method, system, electronic equipment and storage medium | |
| CN111160929A (en) | Method and device for determining client type | |
| CN112580348B (en) | Policy text relevance analysis method and system | |
| CN115080732A (en) | Complaint work order processing method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 1208931 Country of ref document: HK |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |